Trends & Viral SBS Sports · 2026-11-02

AI Is Improving Itself Faster Than Expected, Driving Over 26% of Its Own R&D

Key Points
  • Anthropic revealed that Claude AI now directly drives 26% of its next-generation research and development tasks, up from just 1% a year ago.
  • Tech leaders are sounding alarms that recursive self-improvement could arrive as early as 2027, raising fears over losing control of autonomous development.
  • The news is drawing intense interest in South Korea, where rapid AI integration in workplace tech has made automation and machine control hot-button societal issues.
Summary The rapid acceleration of artificial intelligence has sparked intense global debate, and in South Korea—one of the world's most tech-saturated societies—reports on AI advancing faster than expected are capturing widespread attention. According to recent reporting highlighted by Korean media from Wall Street Journal columnist Ben Cohen, artificial intelligence is approaching "recursive self-improvement" (RSI)—a milestone where AI systems independently iterate and improve upon themselves—far quicker than tech leaders initially anticipated. Data tracked by AI developer Anthropic illustrates just how fast this shift is taking place. By deploying Claude AI agents to analyze over 15,000 internal Slack messages and research tasks, Anthropic discovered that the proportion of research and development executed primarily by AI under human supervision surged dramatically. Just a year ago, AI handled only about 1% of such tasks; by August, Claude was driving a staggering 26% of next-generation Claude R&D. Commentators note that at this trajectory, reaching full automation where humans are largely sidelined from the development loop may solely depend on compute capacity, time, and human policy decisions. Tech executives themselves have voiced deep unease at the velocity of this evolution. Anthropic co-founder Jack Clark shared at a recent International Monetary Fund (IMF) address that upon returning from parental leave earlier this year, he was shocked by engineers' total reliance on AI for coding, prompting him to move up his estimate for RSI from late 2028 to 2027. Meanwhile, Anthropic CEO Dario Amodei has urged extreme caution, emphasizing that without rigorous controls, machine capabilities could rapidly outstrip human comprehension. In tech-forward South Korea, where AI adoption across engineering, business, and everyday life is among the fastest worldwide, these developments resonate deeply, stirring discussions over ethics, safety, and the future of human labor in an increasingly automated era.
Sponsored · Ad